[Debugger] Is "run to here" broken?

On Mon, Nov 13, 2017 at 8:40 PM, Thomas Dupriez <[hidden email]> wrote:

I dug a bit in this issue. Here are the results:


# Problem raised by Stephanne
Code like the following is open in the debugger:

    `myMethod
        1+1. <PROGRAMCOUNTER>
        MyClass new myEvalBlock: [
            2+2. <CURSOR>
        ].
        3+3.`

The program counter is on the 1+1, the cursor is at the end of the 2+2 line.
Right-click, "Run to here".
-> the program counter moves to the 3+3, and does not stop at the 2+2 (where the cursor is). Even though the 2+2 does get evaluated (the myEvalBlock method evaluates the blocks it receives).
If there was no code after the block, the debugger would jump back to the caller of myMethod.

# Why this happens
The implementation of RunToHere (source code below) is basically to stepOver until the context is different or the source code position of the program counter is higher or equal to the source code position of the cursor/selection.

Since the debugger uses stepOver, it executes myEvalBlock without stopping, reaches the 3+3 and see it has gone further than the source code position of the cursorm so it stops.

Source code of DebugSession>>#runToSelection:inContext:

    `runToSelection: selectionInterval inContext: aContext
        "Attempt to step over instructions in selectedContext until the
        execution reaches the selected instruction. This happens when the
        program counter passes the begining of selectionInterval.

        A not nill and valid interval is expected."

        (self pcRangeForContext: aContext) first >= selectionInterval first
            ifTrue: [ ^self ].
        self stepOver: aContext.
        [ aContext == self interruptedContext and: [ (self pcRangeForContext: aContext) first < selectionInterval first ] ]
            whileTrue: [ self stepOver: aContext ]`

# Observations and thoughts
- Replacing the stepOver with a stepInto -> This made the RunToHere stops when resolving the 'new' message (because the context changes).

Thanks for looking into this Thomas.

What happens if you use stepThrough rather than stepInto?

cheers -ben

- Replacing the stepOver with stepInto and removing the equal condition on contexts -> The RunToHere goes to the 3+3. Looking at the source code position of the program counter, it doesn't enter the block and seems to resolve it in a single step. I don't really get why that is, considering using the stepInto button of the debugger does enter the block. Here is the series of source code positions of the program counter during the RunToHere: (21 to: 22)(34 to: 36)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)(65 to: 66).
However, removing the equal condition on contexts means that if the method call returns before reaching the cursor, it won't stop!

- An idea could be to have the RunToHere place a metalink on the selected node and let the execution run until it hits the metalink, which then updates the debugger. Potential problems are that it implies installing a metalink on a method that is already on the stack, which may not be that easy to do properly (in particular, it affects the program counter since it changes the bytecode), and there is the potential case where the metalink is never reached (for example imagine the myEvalBlock: method of my example is just storing the block and not evaluating it).


Cheers,
Thomas

Le 09/11/2017 à 22:06, Stephane Ducasse a écrit :

Agreed. Thomas? It would be a good bone to ... (Un bon os a ronger) .

Stef

On Thu, Nov 9, 2017 at 4:32 AM, Tudor Girba <[hidden email]> wrote:

Hi,

The basic tools, such as debugger, are expected to work. If something does not work, it’s a bug.

Cheers,
Doru

On Nov 8, 2017, at 11:59 PM, Tim Mackinnon [hidden email] wrote:

I think it's broken in Pharo 6 too, as I often find it unreliable.

It's hard to know what should work anymore - we really need a stabilisation release to let the dust settle.

I'm always a bit reticent to report things as I'm not sure what you expect to work.

Tim

Sent from my iPhone

On 8 Nov 2017, at 20:40, Stephane Ducasse <[hidden email]> wrote:

Hi

I have the following method and I have my cursor -MY CURSOR HERE-
I select the menu run to here and .... I exit the method.
:(

Is run to here working in Pharo 70?
I start to get worry about the number of bugs I get when using Pharo70.

Stef


fileOut
"File out the receiver, to a file whose name is a function of the
change-set name and a unique numeric tag."

| nameToUse |
self halt.
self class promptForDefaultChangeSetDirectoryIfNecessary.
nameToUse := (self defaultChangeSetDirectory / self name , 'cs')
nextVersion basename.
UIManager default
showWaitCursorWhile:
[
| internalStream |
internalStream := (String new: 10000) writeStream.

-MY CURSOR HERE-

internalStream
header;
timeStamp.
self fileOutPreambleOn: internalStream.
self fileOutOn: internalStream.
self fileOutPostscriptOn: internalStream.
CodeExporter
writeSourceCodeFrom: internalStream
baseName: (nameToUse copyFrom: 1 to: nameToUse size - 3)
isSt: false ]

--
www.tudorgirba.com
www.feenk.com

"Value is always contextual."

On Mon, Nov 13, 2017 at 9:32 PM, Thomas Dupriez <[hidden email]> wrote:

Le 13/11/2017 à 14:08, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 8:40 PM, Thomas Dupriez <[hidden email]> wrote:

I dug a bit in this issue. Here are the results:


# Problem raised by Stephanne
Code like the following is open in the debugger:

    `myMethod
        1+1. <PROGRAMCOUNTER>
        MyClass new myEvalBlock: [
            2+2. <CURSOR>
        ].
        3+3.`

The program counter is on the 1+1, the cursor is at the end of the 2+2 line.
Right-click, "Run to here".
-> the program counter moves to the 3+3, and does not stop at the 2+2 (where the cursor is). Even though the 2+2 does get evaluated (the myEvalBlock method evaluates the blocks it receives).
If there was no code after the block, the debugger would jump back to the caller of myMethod.

# Why this happens
The implementation of RunToHere (source code below) is basically to stepOver until the context is different or the source code position of the program counter is higher or equal to the source code position of the cursor/selection.

Since the debugger uses stepOver, it executes myEvalBlock without stopping, reaches the 3+3 and see it has gone further than the source code position of the cursorm so it stops.

Source code of DebugSession>>#runToSelection:inContext:

    `runToSelection: selectionInterval inContext: aContext
        "Attempt to step over instructions in selectedContext until the
        execution reaches the selected instruction. This happens when the
        program counter passes the begining of selectionInterval.

        A not nill and valid interval is expected."

        (self pcRangeForContext: aContext) first >= selectionInterval first
            ifTrue: [ ^self ].
        self stepOver: aContext.
        [ aContext == self interruptedContext and: [ (self pcRangeForContext: aContext) first < selectionInterval first ] ]
            whileTrue: [ self stepOver: aContext ]`

# Observations and thoughts
- Replacing the stepOver with a stepInto -> This made the RunToHere stops when resolving the 'new' message (because the context changes).

Thanks for looking into this Thomas.

What happens if you use stepThrough rather than stepInto?

cheers -ben

Using `stepThrough: aContext` instead of `stepInto: aContext`, the RunToHere in the block stops at the intended place: the 2+2.
Printing the sequence of `self pcRangeForContext: aContext` yields the following. It doesn't show precisely the stop in the block, and instead shows an interval encompassing the whole block. That's what I was using to see where the execution was going during the RunToHere loop so I guess that's not a precise enough indicator for this situation...
(21 to: 22)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)

I also tried the case where the block does not get evaluated (changing myEvalBlock so that it does nothing). In this situation, the RunToHere with a stepThrough ends up at the intended place, the 3+3.

So... just use stepThrough for the RunToHere I guess?

Seems like expected behaviour.  

Could you create an Issue/Pull Request to provide some concrete code to review?

Now the trick will be if you can devise some tests to go with this.  

For examples perhaps look at senders of newDebugSessionNamed:startedAt:
including SpecDebugger>>testBasic.

cheers -ben

 

Thomas

 

- Replacing the stepOver with stepInto and removing the equal condition on contexts -> The RunToHere goes to the 3+3. Looking at the source code position of the program counter, it doesn't enter the block and seems to resolve it in a single step. I don't really get why that is, considering using the stepInto button of the debugger does enter the block. Here is the series of source code positions of the program counter during the RunToHere: (21 to: 22)(34 to: 36)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)(65 to: 66).
However, removing the equal condition on contexts means that if the method call returns before reaching the cursor, it won't stop!

- An idea could be to have the RunToHere place a metalink on the selected node and let the execution run until it hits the metalink, which then updates the debugger. Potential problems are that it implies installing a metalink on a method that is already on the stack, which may not be that easy to do properly (in particular, it affects the program counter since it changes the bytecode), and there is the potential case where the metalink is never reached (for example imagine the myEvalBlock: method of my example is just storing the block and not evaluating it).


Cheers,
Thomas

Le 09/11/2017 à 22:06, Stephane Ducasse a écrit :

Agreed. Thomas? It would be a good bone to ... (Un bon os a ronger) .

Stef

On Thu, Nov 9, 2017 at 4:32 AM, Tudor Girba <[hidden email]> wrote:

Hi,

The basic tools, such as debugger, are expected to work. If something does not work, it’s a bug.

Cheers,
Doru

On Nov 8, 2017, at 11:59 PM, Tim Mackinnon [hidden email] wrote:

I think it's broken in Pharo 6 too, as I often find it unreliable.

It's hard to know what should work anymore - we really need a stabilisation release to let the dust settle.

I'm always a bit reticent to report things as I'm not sure what you expect to work.

Tim

Sent from my iPhone

On 8 Nov 2017, at 20:40, Stephane Ducasse <[hidden email]> wrote:

Hi

I have the following method and I have my cursor -MY CURSOR HERE-
I select the menu run to here and .... I exit the method.
:(

Is run to here working in Pharo 70?
I start to get worry about the number of bugs I get when using Pharo70.

Stef


fileOut
"File out the receiver, to a file whose name is a function of the
change-set name and a unique numeric tag."

| nameToUse |
self halt.
self class promptForDefaultChangeSetDirectoryIfNecessary.
nameToUse := (self defaultChangeSetDirectory / self name , 'cs')
nextVersion basename.
UIManager default
showWaitCursorWhile:
[
| internalStream |
internalStream := (String new: 10000) writeStream.

-MY CURSOR HERE-

internalStream
header;
timeStamp.
self fileOutPreambleOn: internalStream.
self fileOutOn: internalStream.
self fileOutPostscriptOn: internalStream.
CodeExporter
writeSourceCodeFrom: internalStream
baseName: (nameToUse copyFrom: 1 to: nameToUse size - 3)
isSt: false ]

--
www.tudorgirba.com
www.feenk.com

"Value is always contextual."

DebugSession-runToSelectioninContext.st (1K) Download Attachment
MyClass.st (940 bytes) Download Attachment

While working on writing a test, I discovered that just using stepThrough instead of stepOver may not be the definitive solution.

# Example1: When using the `value` message directly (instead of calling a method that does that as in my previous example), the RunToHere moves the debugSession to the `value` message and not the inside of the block.

`meth2
    self halt.
    [
        1+1.
        2+2."< cursor before the RunToHere"
    ] ">"value"<result of the RunToHere".
    3+3.
    "Two stepInto are required after the RunToHere to reach the cursor (the `+2` message)"`

# Example2: Going back to using a dedicated method for evaluating the block, but increasing the block's size. The RunToHere moves the debugSession to the first message of the block instead of the pointed one.

`meth3
    self halt.
    MyClass new evalBlock:
    [
        1">"+1"< result of the RunToHere".
        2+2."< cursor before the RunToHere"
    ].
    3+3.
    "One stepInto is required after the RunToHere to reach the cursor (the `+2` message)"`

cheers -thomas

Le 13/11/2017 à 15:56, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 9:32 PM, Thomas Dupriez <[hidden email]> wrote:

Le 13/11/2017 à 14:08, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 8:40 PM, Thomas Dupriez <[hidden email]> wrote:

I dug a bit in this issue. Here are the results:


# Problem raised by Stephanne
Code like the following is open in the debugger:

    `myMethod
        1+1. <PROGRAMCOUNTER>
        MyClass new myEvalBlock: [
            2+2. <CURSOR>
        ].
        3+3.`

The program counter is on the 1+1, the cursor is at the end of the 2+2 line.
Right-click, "Run to here".
-> the program counter moves to the 3+3, and does not stop at the 2+2 (where the cursor is). Even though the 2+2 does get evaluated (the myEvalBlock method evaluates the blocks it receives).
If there was no code after the block, the debugger would jump back to the caller of myMethod.

# Why this happens
The implementation of RunToHere (source code below) is basically to stepOver until the context is different or the source code position of the program counter is higher or equal to the source code position of the cursor/selection.

Since the debugger uses stepOver, it executes myEvalBlock without stopping, reaches the 3+3 and see it has gone further than the source code position of the cursorm so it stops.

Source code of DebugSession>>#runToSelection:inContext:

    `runToSelection: selectionInterval inContext: aContext
        "Attempt to step over instructions in selectedContext until the
        execution reaches the selected instruction. This happens when the
        program counter passes the begining of selectionInterval.

        A not nill and valid interval is expected."

        (self pcRangeForContext: aContext) first >= selectionInterval first
            ifTrue: [ ^self ].
        self stepOver: aContext.
        [ aContext == self interruptedContext and: [ (self pcRangeForContext: aContext) first < selectionInterval first ] ]
            whileTrue: [ self stepOver: aContext ]`

# Observations and thoughts
- Replacing the stepOver with a stepInto -> This made the RunToHere stops when resolving the 'new' message (because the context changes).

Thanks for looking into this Thomas.

What happens if you use stepThrough rather than stepInto?

cheers -ben

Using `stepThrough: aContext` instead of `stepInto: aContext`, the RunToHere in the block stops at the intended place: the 2+2.
Printing the sequence of `self pcRangeForContext: aContext` yields the following. It doesn't show precisely the stop in the block, and instead shows an interval encompassing the whole block. That's what I was using to see where the execution was going during the RunToHere loop so I guess that's not a precise enough indicator for this situation...
(21 to: 22)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)

I also tried the case where the block does not get evaluated (changing myEvalBlock so that it does nothing). In this situation, the RunToHere with a stepThrough ends up at the intended place, the 3+3.

So... just use stepThrough for the RunToHere I guess?

Seems like expected behaviour.  

Could you create an Issue/Pull Request to provide some concrete code to review?

Now the trick will be if you can devise some tests to go with this.  

For examples perhaps look at senders of newDebugSessionNamed:startedAt:
including SpecDebugger>>testBasic.

cheers -ben

 

I opened an issue on FogBugz: https://pharo.fogbugz.com/f/cases/20687/

Attached are the file out of the test code (MyClass.st) and the fixed runToSelection:inContext: (DebugSession-runToSelectioninContext).
I'll make a pull request tomorrow.

For the test, I'm thinking about something like the following, that checks the source code position of the pc after the runToSelection. (It's not working yet)
Now that I think of it, the source code position of the pc will maybe be the whole block (as experienced earlier) so this approach may end up not working.
I'll come back to that tomorrow.

Draft for a test:
    `testRunToSelectionInContext
    |context_ process_ session_|
    "The variable names have an underscore to distinguish them from those declared in the setUp method of this TestCase (DebuggerModelTest), which are not suited for this test (because not making a debug session out of suitable code)"
    context_ := [1+1.
        self evalBlock:[2+2].
        3+3.
    ] asContext.
    process_ := Process
        forContext: context_
        priority: Processor userInterruptPriority.
       
    session_:= process_ newDebugSessionNamed: 'test session' startedAt: context_.
    session_ runToSelection: (Interval from: 28 to: 27) inContext: context_.
    self assert: ((session_ pcRangeForContext: context_) first == 3)`

cheers - Thomas

Thomas

 

- Replacing the stepOver with stepInto and removing the equal condition on contexts -> The RunToHere goes to the 3+3. Looking at the source code position of the program counter, it doesn't enter the block and seems to resolve it in a single step. I don't really get why that is, considering using the stepInto button of the debugger does enter the block. Here is the series of source code positions of the program counter during the RunToHere: (21 to: 22)(34 to: 36)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)(65 to: 66).
However, removing the equal condition on contexts means that if the method call returns before reaching the cursor, it won't stop!

- An idea could be to have the RunToHere place a metalink on the selected node and let the execution run until it hits the metalink, which then updates the debugger. Potential problems are that it implies installing a metalink on a method that is already on the stack, which may not be that easy to do properly (in particular, it affects the program counter since it changes the bytecode), and there is the potential case where the metalink is never reached (for example imagine the myEvalBlock: method of my example is just storing the block and not evaluating it).


Cheers,
Thomas

Le 09/11/2017 à 22:06, Stephane Ducasse a écrit :

Agreed. Thomas? It would be a good bone to ... (Un bon os a ronger) .

Stef

On Thu, Nov 9, 2017 at 4:32 AM, Tudor Girba <[hidden email]> wrote:

Hi,

The basic tools, such as debugger, are expected to work. If something does not work, it’s a bug.

Cheers,
Doru

On Nov 8, 2017, at 11:59 PM, Tim Mackinnon [hidden email] wrote:

I think it's broken in Pharo 6 too, as I often find it unreliable.

It's hard to know what should work anymore - we really need a stabilisation release to let the dust settle.

I'm always a bit reticent to report things as I'm not sure what you expect to work.

Tim

Sent from my iPhone

On 8 Nov 2017, at 20:40, Stephane Ducasse <[hidden email]> wrote:

Hi

I have the following method and I have my cursor -MY CURSOR HERE-
I select the menu run to here and .... I exit the method.
:(

Is run to here working in Pharo 70?
I start to get worry about the number of bugs I get when using Pharo70.

Stef


fileOut
"File out the receiver, to a file whose name is a function of the
change-set name and a unique numeric tag."

| nameToUse |
self halt.
self class promptForDefaultChangeSetDirectoryIfNecessary.
nameToUse := (self defaultChangeSetDirectory / self name , 'cs')
nextVersion basename.
UIManager default
showWaitCursorWhile:
[
| internalStream |
internalStream := (String new: 10000) writeStream.

-MY CURSOR HERE-

internalStream
header;
timeStamp.
self fileOutPreambleOn: internalStream.
self fileOutOn: internalStream.
self fileOutPostscriptOn: internalStream.
CodeExporter
writeSourceCodeFrom: internalStream
baseName: (nameToUse copyFrom: 1 to: nameToUse size - 3)
isSt: false ]

--
www.tudorgirba.com
www.feenk.com

"Value is always contextual."

On 14 November 2017 at 23:55, Thomas Dupriez <[hidden email]> wrote:

While working on writing a test, I discovered that just using stepThrough instead of stepOver may not be the definitive solution.

# Example1: When using the `value` message directly (instead of calling a method that does that as in my previous example), the RunToHere moves the debugSession to the `value` message and not the inside of the block.

`meth2
    self halt.
    [
        1+1.
        2+2."< cursor before the RunToHere"
    ] ">"value"<result of the RunToHere".
    3+3.
    "Two stepInto are required after the RunToHere to reach the cursor (the `+2` message)"`

# Example2: Going back to using a dedicated method for evaluating the block, but increasing the block's size. The RunToHere moves the debugSession to the first message of the block instead of the pointed one.

`meth3
    self halt.
    MyClass new evalBlock:
    [
        1">"+1"< result of the RunToHere".
        2+2."< cursor before the RunToHere"
    ].
    3+3.
    "One stepInto is required after the RunToHere to reach the cursor (the `+2` message)"`

cheers -thomas

In both cases, what happens with an additional step-Through?

cheers -ben

 

Le 13/11/2017 à 17:23, Thomas Dupriez a écrit :

Le 13/11/2017 à 15:56, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 9:32 PM, Thomas Dupriez <[hidden email]> wrote:

Le 13/11/2017 à 14:08, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 8:40 PM, Thomas Dupriez <[hidden email]> wrote:

I dug a bit in this issue. Here are the results:


# Problem raised by Stephanne
Code like the following is open in the debugger:

    `myMethod
        1+1. <PROGRAMCOUNTER>
        MyClass new myEvalBlock: [
            2+2. <CURSOR>
        ].
        3+3.`

The program counter is on the 1+1, the cursor is at the end of the 2+2 line.
Right-click, "Run to here".
-> the program counter moves to the 3+3, and does not stop at the 2+2 (where the cursor is). Even though the 2+2 does get evaluated (the myEvalBlock method evaluates the blocks it receives).
If there was no code after the block, the debugger would jump back to the caller of myMethod.

# Why this happens
The implementation of RunToHere (source code below) is basically to stepOver until the context is different or the source code position of the program counter is higher or equal to the source code position of the cursor/selection.

Since the debugger uses stepOver, it executes myEvalBlock without stopping, reaches the 3+3 and see it has gone further than the source code position of the cursorm so it stops.

Source code of DebugSession>>#runToSelection:inContext:

    `runToSelection: selectionInterval inContext: aContext
        "Attempt to step over instructions in selectedContext until the
        execution reaches the selected instruction. This happens when the
        program counter passes the begining of selectionInterval.

        A not nill and valid interval is expected."

        (self pcRangeForContext: aContext) first >= selectionInterval first
            ifTrue: [ ^self ].
        self stepOver: aContext.
        [ aContext == self interruptedContext and: [ (self pcRangeForContext: aContext) first < selectionInterval first ] ]
            whileTrue: [ self stepOver: aContext ]`

# Observations and thoughts
- Replacing the stepOver with a stepInto -> This made the RunToHere stops when resolving the 'new' message (because the context changes).

Thanks for looking into this Thomas.

What happens if you use stepThrough rather than stepInto?

cheers -ben

Using `stepThrough: aContext` instead of `stepInto: aContext`, the RunToHere in the block stops at the intended place: the 2+2.
Printing the sequence of `self pcRangeForContext: aContext` yields the following. It doesn't show precisely the stop in the block, and instead shows an interval encompassing the whole block. That's what I was using to see where the execution was going during the RunToHere loop so I guess that's not a precise enough indicator for this situation...
(21 to: 22)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)

I also tried the case where the block does not get evaluated (changing myEvalBlock so that it does nothing). In this situation, the RunToHere with a stepThrough ends up at the intended place, the 3+3.

So... just use stepThrough for the RunToHere I guess?

Seems like expected behaviour.  

Could you create an Issue/Pull Request to provide some concrete code to review?

Now the trick will be if you can devise some tests to go with this.  

For examples perhaps look at senders of newDebugSessionNamed:startedAt:
including SpecDebugger>>testBasic.

cheers -ben

 

I opened an issue on FogBugz: https://pharo.fogbugz.com/f/cases/20687/

Attached are the file out of the test code (MyClass.st) and the fixed runToSelection:inContext: (DebugSession-runToSelectioninContext).
I'll make a pull request tomorrow.

For the test, I'm thinking about something like the following, that checks the source code position of the pc after the runToSelection. (It's not working yet)
Now that I think of it, the source code position of the pc will maybe be the whole block (as experienced earlier) so this approach may end up not working.
I'll come back to that tomorrow.

Draft for a test:
    `testRunToSelectionInContext
    |context_ process_ session_|
    "The variable names have an underscore to distinguish them from those declared in the setUp method of this TestCase (DebuggerModelTest), which are not suited for this test (because not making a debug session out of suitable code)"
    context_ := [1+1.
        self evalBlock:[2+2].
        3+3.
    ] asContext.
    process_ := Process
        forContext: context_
        priority: Processor userInterruptPriority.
       
    session_:= process_ newDebugSessionNamed: 'test session' startedAt: context_.
    session_ runToSelection: (Interval from: 28 to: 27) inContext: context_.
    self assert: ((session_ pcRangeForContext: context_) first == 3)`

cheers - Thomas

Thomas

 

- Replacing the stepOver with stepInto and removing the equal condition on contexts -> The RunToHere goes to the 3+3. Looking at the source code position of the program counter, it doesn't enter the block and seems to resolve it in a single step. I don't really get why that is, considering using the stepInto button of the debugger does enter the block. Here is the series of source code positions of the program counter during the RunToHere: (21 to: 22)(34 to: 36)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)(65 to: 66).
However, removing the equal condition on contexts means that if the method call returns before reaching the cursor, it won't stop!

- An idea could be to have the RunToHere place a metalink on the selected node and let the execution run until it hits the metalink, which then updates the debugger. Potential problems are that it implies installing a metalink on a method that is already on the stack, which may not be that easy to do properly (in particular, it affects the program counter since it changes the bytecode), and there is the potential case where the metalink is never reached (for example imagine the myEvalBlock: method of my example is just storing the block and not evaluating it).


Cheers,
Thomas

Le 09/11/2017 à 22:06, Stephane Ducasse a écrit :

Agreed. Thomas? It would be a good bone to ... (Un bon os a ronger) .

Stef

On Thu, Nov 9, 2017 at 4:32 AM, Tudor Girba <[hidden email]> wrote:

Hi,

The basic tools, such as debugger, are expected to work. If something does not work, it’s a bug.

Cheers,
Doru

On Nov 8, 2017, at 11:59 PM, Tim Mackinnon [hidden email] wrote:

I think it's broken in Pharo 6 too, as I often find it unreliable.

It's hard to know what should work anymore - we really need a stabilisation release to let the dust settle.

I'm always a bit reticent to report things as I'm not sure what you expect to work.

Tim

Sent from my iPhone

On 8 Nov 2017, at 20:40, Stephane Ducasse <[hidden email]> wrote:

Hi

I have the following method and I have my cursor -MY CURSOR HERE-
I select the menu run to here and .... I exit the method.
:(

Is run to here working in Pharo 70?
I start to get worry about the number of bugs I get when using Pharo70.

Stef


fileOut
"File out the receiver, to a file whose name is a function of the
change-set name and a unique numeric tag."

| nameToUse |
self halt.
self class promptForDefaultChangeSetDirectoryIfNecessary.
nameToUse := (self defaultChangeSetDirectory / self name , 'cs')
nextVersion basename.
UIManager default
showWaitCursorWhile:
[
| internalStream |
internalStream := (String new: 10000) writeStream.

-MY CURSOR HERE-

internalStream
header;
timeStamp.
self fileOutPreambleOn: internalStream.
self fileOutOn: internalStream.
self fileOutPostscriptOn: internalStream.
CodeExporter
writeSourceCodeFrom: internalStream
baseName: (nameToUse copyFrom: 1 to: nameToUse size - 3)
isSt: false ]

--
www.tudorgirba.com
www.feenk.com

"Value is always contextual."

Here they are:

https://github.com/dupriezt/Pharo-RunToSelectionDebuggerFeature

Thomas, can you try writing some real test cases in SUnit?

I don't care if they are originally red (they will be because it is not working).

The idea is that we capture all these different scenarios and have a way to know automatically in the future if

 - we fixed all them

 - or we have a regression and broke some.

On Wed, Nov 15, 2017 at 10:44 AM, Thomas Dupriez <[hidden email]> wrote:

Le 15/11/2017 à 01:19, Ben Coman a écrit :

On 14 November 2017 at 23:55, Thomas Dupriez <[hidden email]> wrote:

While working on writing a test, I discovered that just using stepThrough instead of stepOver may not be the definitive solution.

# Example1: When using the `value` message directly (instead of calling a method that does that as in my previous example), the RunToHere moves the debugSession to the `value` message and not the inside of the block.

`meth2
    self halt.
    [
        1+1.
        2+2."< cursor before the RunToHere"
    ] ">"value"<result of the RunToHere".
    3+3.
    "Two stepInto are required after the RunToHere to reach the cursor (the `+2` message)"`

# Example2: Going back to using a dedicated method for evaluating the block, but increasing the block's size. The RunToHere moves the debugSession to the first message of the block instead of the pointed one.

`meth3
    self halt.
    MyClass new evalBlock:
    [
        1">"+1"< result of the RunToHere".
        2+2."< cursor before the RunToHere"
    ].
    3+3.
    "One stepInto is required after the RunToHere to reach the cursor (the `+2` message)"`

cheers -thomas

In both cases, what happens with an additional step-Through?

cheers -ben

 

# Example1*:
Operation: RunToHere with cursor at the end of the `2+2.` line + 1 stepThrough.
Result: DebugSession at the `+1`, first instruction of the block but not the intended position.
Note: Doing an additional stepThrough brings the debugSession to the intended `+2`.

`meth2
    self halt.
    [
        1">"+1"<"."result of the RunToHere + 1 stepThrough"
        2+2."< cursor before the RunToHere"
    ] ">"value"<result of the RunToHere".
    3+3.
    "Two stepInto are required after the RunToHere to reach the cursor (the `+2` message)"`

# Example2*:
Operation: RunToHere with cursor at the end of the `2+2.` line + 1 stepThrough
Result: DebugSession at the intended `+2` position.
Note: The additional stepThrough just moves the DebugSession to the next instruction. See Example3 for what happens when the target instruction is further away.

`meth3
    self halt.
    MyClass new evalBlock:
    [
        1">"+1"< result of the RunToHere".
        2">"+2"<result of RunToHere+1stepThrough"."< cursor before the RunToHere"
    ].
    3+3.
    "One stepInto is required after the RunToHere to reach the cursor (the `+2` message)"`


# Example3:
Operation: RunToHere with cursor at the end of the `3+3.` line. + 1 stepThrough.
Result: DebugSession is at the `+2` instruction, not the intended `+3` one.

`meth4
    self halt.
    MyClass new evalBlock:
    [
        1">"+1"< result of the RunToHere".
        2">"+2"<result of RunToHere+1stepThrough".
        3">"+3"<result of RunToHere+2stepThrough"."< cursor before the RunToHere"
    ].
    4+4.
    "One stepInto is required after the RunToHere to reach the cursor (the `+2` message)"`

cheers -thomas

Le 13/11/2017 à 17:23, Thomas Dupriez a écrit :

Le 13/11/2017 à 15:56, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 9:32 PM, Thomas Dupriez <[hidden email]> wrote:

Le 13/11/2017 à 14:08, Ben Coman a écrit :

On Mon, Nov 13, 2017 at 8:40 PM, Thomas Dupriez <[hidden email]> wrote:

I dug a bit in this issue. Here are the results:


# Problem raised by Stephanne
Code like the following is open in the debugger:

    `myMethod
        1+1. <PROGRAMCOUNTER>
        MyClass new myEvalBlock: [
            2+2. <CURSOR>
        ].
        3+3.`

The program counter is on the 1+1, the cursor is at the end of the 2+2 line.
Right-click, "Run to here".
-> the program counter moves to the 3+3, and does not stop at the 2+2 (where the cursor is). Even though the 2+2 does get evaluated (the myEvalBlock method evaluates the blocks it receives).
If there was no code after the block, the debugger would jump back to the caller of myMethod.

# Why this happens
The implementation of RunToHere (source code below) is basically to stepOver until the context is different or the source code position of the program counter is higher or equal to the source code position of the cursor/selection.

Since the debugger uses stepOver, it executes myEvalBlock without stopping, reaches the 3+3 and see it has gone further than the source code position of the cursorm so it stops.

Source code of DebugSession>>#runToSelection:inContext:

    `runToSelection: selectionInterval inContext: aContext
        "Attempt to step over instructions in selectedContext until the
        execution reaches the selected instruction. This happens when the
        program counter passes the begining of selectionInterval.

        A not nill and valid interval is expected."

        (self pcRangeForContext: aContext) first >= selectionInterval first
            ifTrue: [ ^self ].
        self stepOver: aContext.
        [ aContext == self interruptedContext and: [ (self pcRangeForContext: aContext) first < selectionInterval first ] ]
            whileTrue: [ self stepOver: aContext ]`

# Observations and thoughts
- Replacing the stepOver with a stepInto -> This made the RunToHere stops when resolving the 'new' message (because the context changes).

Thanks for looking into this Thomas.

What happens if you use stepThrough rather than stepInto?

cheers -ben

Using `stepThrough: aContext` instead of `stepInto: aContext`, the RunToHere in the block stops at the intended place: the 2+2.
Printing the sequence of `self pcRangeForContext: aContext` yields the following. It doesn't show precisely the stop in the block, and instead shows an interval encompassing the whole block. That's what I was using to see where the execution was going during the RunToHere loop so I guess that's not a precise enough indicator for this situation...
(21 to: 22)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)

I also tried the case where the block does not get evaluated (changing myEvalBlock so that it does nothing). In this situation, the RunToHere with a stepThrough ends up at the intended place, the 3+3.

So... just use stepThrough for the RunToHere I guess?

Seems like expected behaviour.  

Could you create an Issue/Pull Request to provide some concrete code to review?

Now the trick will be if you can devise some tests to go with this.  

For examples perhaps look at senders of newDebugSessionNamed:startedAt:
including SpecDebugger>>testBasic.

cheers -ben

 

I opened an issue on FogBugz: https://pharo.fogbugz.com/f/cases/20687/

Attached are the file out of the test code (MyClass.st) and the fixed runToSelection:inContext: (DebugSession-runToSelectioninContext).
I'll make a pull request tomorrow.

For the test, I'm thinking about something like the following, that checks the source code position of the pc after the runToSelection. (It's not working yet)
Now that I think of it, the source code position of the pc will maybe be the whole block (as experienced earlier) so this approach may end up not working.
I'll come back to that tomorrow.

Draft for a test:
    `testRunToSelectionInContext
    |context_ process_ session_|
    "The variable names have an underscore to distinguish them from those declared in the setUp method of this TestCase (DebuggerModelTest), which are not suited for this test (because not making a debug session out of suitable code)"
    context_ := [1+1.
        self evalBlock:[2+2].
        3+3.
    ] asContext.
    process_ := Process
        forContext: context_
        priority: Processor userInterruptPriority.
       
    session_:= process_ newDebugSessionNamed: 'test session' startedAt: context_.
    session_ runToSelection: (Interval from: 28 to: 27) inContext: context_.
    self assert: ((session_ pcRangeForContext: context_) first == 3)`

cheers - Thomas

Thomas

 

- Replacing the stepOver with stepInto and removing the equal condition on contexts -> The RunToHere goes to the 3+3. Looking at the source code position of the program counter, it doesn't enter the block and seems to resolve it in a single step. I don't really get why that is, considering using the stepInto button of the debugger does enter the block. Here is the series of source code positions of the program counter during the RunToHere: (21 to: 22)(34 to: 36)(34 to: 36)(49 to: 60)(38 to: 60)(38 to: 60)(65 to: 66).
However, removing the equal condition on contexts means that if the method call returns before reaching the cursor, it won't stop!

- An idea could be to have the RunToHere place a metalink on the selected node and let the execution run until it hits the metalink, which then updates the debugger. Potential problems are that it implies installing a metalink on a method that is already on the stack, which may not be that easy to do properly (in particular, it affects the program counter since it changes the bytecode), and there is the potential case where the metalink is never reached (for example imagine the myEvalBlock: method of my example is just storing the block and not evaluating it).


Cheers,
Thomas

Le 09/11/2017 à 22:06, Stephane Ducasse a écrit :

Agreed. Thomas? It would be a good bone to ... (Un bon os a ronger) .

Stef

On Thu, Nov 9, 2017 at 4:32 AM, Tudor Girba <[hidden email]> wrote:

Hi,

The basic tools, such as debugger, are expected to work. If something does not work, it’s a bug.

Cheers,
Doru

On Nov 8, 2017, at 11:59 PM, Tim Mackinnon [hidden email] wrote:

I think it's broken in Pharo 6 too, as I often find it unreliable.

It's hard to know what should work anymore - we really need a stabilisation release to let the dust settle.

I'm always a bit reticent to report things as I'm not sure what you expect to work.

Tim

Sent from my iPhone

On 8 Nov 2017, at 20:40, Stephane Ducasse <[hidden email]> wrote:

Hi

I have the following method and I have my cursor -MY CURSOR HERE-
I select the menu run to here and .... I exit the method.
:(

Is run to here working in Pharo 70?
I start to get worry about the number of bugs I get when using Pharo70.

Stef


fileOut
"File out the receiver, to a file whose name is a function of the
change-set name and a unique numeric tag."

| nameToUse |
self halt.
self class promptForDefaultChangeSetDirectoryIfNecessary.
nameToUse := (self defaultChangeSetDirectory / self name , 'cs')
nextVersion basename.
UIManager default
showWaitCursorWhile:
[
| internalStream |
internalStream := (String new: 10000) writeStream.

-MY CURSOR HERE-

internalStream
header;
timeStamp.
self fileOutPreambleOn: internalStream.
self fileOutOn: internalStream.
self fileOutPostscriptOn: internalStream.
CodeExporter
writeSourceCodeFrom: internalStream
baseName: (nameToUse copyFrom: 1 to: nameToUse size - 3)
isSt: false ]

--
www.tudorgirba.com
www.feenk.com

"Value is always contextual."

--

Guille Polito Research Engineer Centre de Recherche en Informatique, Signal et Automatique de Lille CRIStAL - UMR 9189 French National Center for Scientific Research - http://www.cnrs.fr Web: http://guillep.github.io Phone: +33 06 52 70 66 13